Lab 11 - Interactive Visualization

Learning Goals

  • Read in and process the COVID dataset from the New York Times GitHub repository
  • Create interactive graphs of different types using plot_ly() and ggplotly() functions
  • Customize the hoverinfo and other plot features
  • Create a Choropleth map using plot_geo()

Lab Description

We will work with COVID data downloaded from the New York Times. The dataset consists of COVID-19 cases and deaths in each US state during the course of the COVID epidemic.

The objective of this lab is to explore relationships between cases, deaths, and population sizes of US states, and plot data to demonstrate this

Steps

I. Reading and processing the New York Times (NYT) state-level COVID-19 data

0. Install and load libraries

library(data.table)
library(tidyverse)
## ── Attaching packages ─────────────────────────────────────── tidyverse 1.3.2 ──
## ✔ ggplot2 3.4.0      ✔ purrr   1.0.1 
## ✔ tibble  3.1.8      ✔ dplyr   1.0.10
## ✔ tidyr   1.2.1      ✔ stringr 1.5.0 
## ✔ readr   2.1.3      ✔ forcats 0.5.2 
## ── Conflicts ────────────────────────────────────────── tidyverse_conflicts() ──
## ✖ dplyr::between()   masks data.table::between()
## ✖ dplyr::filter()    masks stats::filter()
## ✖ dplyr::first()     masks data.table::first()
## ✖ dplyr::lag()       masks stats::lag()
## ✖ dplyr::last()      masks data.table::last()
## ✖ purrr::transpose() masks data.table::transpose()
library(plotly)
## 
## Attaching package: 'plotly'
## 
## The following object is masked from 'package:ggplot2':
## 
##     last_plot
## 
## The following object is masked from 'package:stats':
## 
##     filter
## 
## The following object is masked from 'package:graphics':
## 
##     layout
library(knitr)
library(widgetframe)
## Loading required package: htmlwidgets

1. Read in the data

cv_states_readin <- 
  data.table::fread("https://raw.githubusercontent.com/nytimes/covid-19-data/master/us-states.csv")


state_pops <- data.table::fread("https://raw.githubusercontent.com/COVID19Tracking/associated-data/master/us_census_data/us_census_2018_population_estimates_states.csv")

state_pops$abb <- state_pops$state
state_pops$state <- state_pops$state_name
state_pops$state_name <- NULL

cv_states <- merge(cv_states_readin, state_pops, by="state")

2. Look at the data

  • Inspect the dimensions, head, and tail of the data
  • Inspect the structure of each variables. Are they in the correct format?
dim(cv_states)
head(cv_states)
tail(cv_states)
str(cv_states)

3. Format the data

  • Make date into a date variable
  • Make state into a factor variable
  • Order the data first by state, second by date
  • Confirm the variables are now correctly formatted
  • Inspect the range values for each variable. What is the date range? The range of cases and deaths?
cv_states$date <- as.Date(cv_states$date, format="%Y-%m-%d")


state_list <- unique(cv_states$state)
cv_states$state <- factor(cv_states$state, levels = state_list)
abb_list <- unique(cv_states$abb)
cv_states$abb <- factor(cv_states$abb, levels = abb_list)


cv_states = cv_states[order(cv_states$state, cv_states$date),]
str(cv_states)
head(cv_states)
tail(cv_states)


head(cv_states)
summary(cv_states)
min(cv_states$date)
max(cv_states$date)

Answer: The date range is from 2020-03-13 to 2023-03-21. The cases range is 1 to 12,154,941, and the deaths range is 0 to 104,185 .

4. Add new_cases and new_deaths and correct outliers

  • Add variables for new cases, new_cases, and new deaths, new_deaths:
    • Hint: You can set new_cases equal to the difference between cases on date i and date i-1, starting on date i=2
new_cv_states <- cv_states %>% 
  group_by(state) %>% 
  mutate(
    new_cases = c(-9999, diff(cases)),
    new_deaths = c(-9999, diff(deaths))
  ) %>% 
  mutate(
    new_cases = ifelse(new_cases == -9999, cases, new_cases),
    new_deaths = ifelse(new_deaths == -9999, deaths, new_deaths)
  )
  • Filter to dates after October 1, 2022
time_cv_states <- new_cv_states %>% filter(date >= '2022-10-01')
  • Use ggplotly for EDA: See if there are outliers or values that don’t make sense for new_cases and new_deaths. Which states and which dates have strange values?
plt_case <- ggplot(
  time_cv_states, aes(x = date, y = new_cases, colour = state)) + 
  geom_line() +
  theme_bw() + 
  ggtitle("New cases in the US since 2022-10-01")
ggplotly(plt_case)
plt_death <- ggplot(
  time_cv_states, aes(x = date, y = new_deaths, colour = state)) + 
  geom_line() +
  theme_bw() + 
  ggtitle("New deaths in the US since 2022-10-01")
ggplotly(plt_death)

Answer: There are a few negative values for both new_cases and new deaths, which is definitely not possible. For example, Tennesse on 2023-01-01, this is likely due to errors in data. Also, new york has some extremely high new deaths on 2022-11-11, which seems abnormal.

  • Correct outliers: Set negative values for new_cases or new_deaths to 0

  • Inspect data again interactively

time_cv_states <- time_cv_states %>% mutate(
  new_cases = ifelse(new_cases < 0, 0, new_cases),
  new_deaths = ifelse(new_deaths < 0, 0, new_deaths)
)

5. Add additional variables

  • Add population-normalized (by 100,000) variables for each variable type (rounded to 1 decimal place). Make sure the variables you calculate are in the correct format (numeric). You can use the following variable names:

    • per100k = cases per 100,000 population
    • newper100k= new cases per 100,000
    • deathsper100k = deaths per 100,000
    • newdeathsper100k = new deaths per 100,000
  • Add a “naive CFR” variable representing deaths / cases on each date for each state

  • Create a dataframe representing values on the most recent date, cv_states_today

per100_cv_states <- time_cv_states %>% 
  mutate(
    per100k = round(cases / population * 1e5, digits = 1),
    newper100k = round(new_cases / population * 1e5, digits = 1),
    deathsper100k = round(deaths / population * 1e5, digits = 1),
    newdeathsper100k = round(new_deaths / population * 1e5, digits = 1)
  )
cfr_states <- per100_cv_states %>% mutate(
  naive_cfr = deaths / cases
)
max_date <- max(cfr_states$date)
cv_states_today <- cfr_states %>% filter(date == max_date)

II. Scatterplots

6. Explore scatterplots using plot_ly()

  • Create a scatterplot using plot_ly() representing pop_density vs. various variables (e.g. cases, per100k, deaths, deathsper100k) for each state on most recent date (cv_states_today)
    • Color points by state and size points by state population
    • Use hover to identify any outliers.
plot_ly(
  cv_states_today,
  x = ~log(pop_density),
  y = ~per100k,
  color = ~state,
  size = ~population,
  type = "scatter",
  sizes = c(5, 100),
  marker = list(sizemode = "diameter", opacity = 0.8)
) %>% layout(title = "Cases per 100k vs state population ")
  • Remove those outliers and replot.
# Removing District of Columbia
plot_ly(
  cv_states_today %>% filter(state != "District of Columbia"),
  x = ~pop_density,
  y = ~per100k,
  color = ~state,
  size = ~population,
  type = "scatter",
  sizes = c(5, 100),
  marker = list(sizemode = "diameter", opacity = 0.8)
) %>% layout(title = "Cases vs state population without District of Columbia")
  • Choose one plot. For this plot:
  • Add hoverinfo specifying the state name, cases per 100k, and deaths per 100k, similarly to how we did this in the lecture notes
  • Add layout information to title the chart and the axes
  • Enable hovermode = "compare"
plot_ly(
  cv_states_today,
  x = ~log(pop_density),
  y = ~cases,
  color = ~state,
  size = ~population,
  type = "scatter",
  sizes = c(5, 50),
  marker = list(sizemode = "diameter", 
                opacity = 0.8),
  hoverinfo = "text",
  text = ~ paste0(state, "\n", " Cases per 100k: ", per100k, "\n",
                " Deaths per 100k: ", deathsper100k, "\n",
                " Population density: ", round(pop_density, 1), " per sq miles")
) %>% layout(title = "Cumulative cases vs log population density",
             hovermode = 'x')

7. Explore scatterplot trend interactively using ggplotly() and geom_smooth()

  • For pop_density vs. newdeathsper100k create a chart with the same variables using gglot_ly()
  • Explore the pattern between \(x\) and \(y\)
  • Explain what you see. Do you think pop_density correlates with newdeathsper100k?
plt_smooth <- ggplot(
  cv_states_today, 
  aes(x = pop_density, y = newdeathsper100k)) +
  geom_point(aes(color = state, size = population)) + 
  geom_smooth() +
  theme_minimal() +
  scale_x_continuous(trans = "log") + 
  ylab("new deaths per 100k") + 
  xlab("population density")

ggplotly(plt_smooth)

Answer: It seems like that population density isn’t correlated with new deaths per 1000 population, since the fitted line is almost flat, and we don’t see any distinct trends in the points as well.

8. Multiple line chart

  • Create a line chart of the naive_CFR for all states over time using plot_ly()
    • Use the zoom and pan tools to inspect the naive_CFR for the states that had an increase in September. How have they changed over time?
plot_ly(
  cfr_states, 
  x = ~date,
  y = ~naive_cfr, 
  color = ~ state, 
  mode = "lines"
) %>% layout(hovermode = "x unified") # added for better comparison
  • Create one more line chart, for Florida only, which shows new_cases and new_deaths together in one plot. Hint: use add_layer()
# I think you were just missing a type = "scatter" for it to work
plot_ly(
  cfr_states %>% filter(state == "Florida"),
  x = ~ date,
  y = ~ new_cases,
  type = "scatter",
  mode = "linear",
  name = "cases"
) %>% add_trace(
  y = ~new_deaths,
  name = "deaths")
  • Use hoverinfo to “eyeball” the approximate peak of deaths and peak of cases. What is the time delay between the peak of cases and the peak of deaths?

Answer: It seems that the peak of deaths and peak of cases has a lag of 0. For example there’s a peak of cases on Jan 6th, and a peak of deaths of 264 deaths on Jan 6th as well. Almost all the peaks of cases and deaths match.

9. Heatmaps

Create a heatmap to visualize new_cases for each state on each date greater than January 1st, 2023 - Start by mapping selected features in the dataframe into a matrix using the tidyr package function pivot_wider(), naming the rows and columns, as done in the lecture notes - Use plot_ly() to create a heatmap out of this matrix. Which states stand out?

cv_states_mat <- cfr_states %>%  select(state, date, new_cases) %>% 
  pivot_wider(names_from = state, values_from = new_cases) %>% 
  column_to_rownames("date")
cv_states_mat %>% 
  plot_ly(
    x = rownames(cv_states_mat),
    y = colnames(cv_states_mat),
    z = ~as.matrix(cv_states_mat),
    type = "heatmap",
    showscale = TRUE
  )

Answer: It seems like Virginia, Washington, North Dakata had very high new cases, and a lot of them is on the date Oct 5, 2022.

  • Create a second heatmap in which the pattern of new_cases for each state over time becomes more clear by filtering to only look at dates every two weeks.
#create heatmap
filter_dates <- seq(
  as.Date("2022-10-01"),
  max_date,
  by = "3 days"
)

cv_states_mat2 <- cfr_states %>%  select(state, date, new_cases) %>% 
  filter(date %in% filter_dates) %>% 
  pivot_wider(names_from = state, values_from = new_cases) %>% 
  column_to_rownames("date")

cv_states_mat2 %>% 
  plot_ly(
    x = rownames(cv_states_mat2),
    y = colnames(cv_states_mat2),
    z = ~as.matrix(cv_states_mat2),
    type = "heatmap",
    showscale = TRUE
  )

10. Map

  • Create a map to visualize the naive_CFR by state on March 15, 2023
pick.date = "2023-03-15"

# Create the map
cv_march_15 <- cfr_states %>% 
  filter(date == pick.date)

plot_geo(cv_march_15, 
         locationmode = "USA-states") %>% 
  add_trace(
    z = ~naive_cfr,
    locations = ~abb
  ) %>% 
  layout(
    geo = list(
      scope = "usa",
      showlakes = TRUE,
      lakecolor = toRGB("red")
    )
  )
  • Compare with a map visualizing the naive_CFR by state on most recent date
# Map for today's date

plot_geo(cv_states_today, 
         locationmode = "USA-states") %>% 
  add_trace(
    z = ~naive_cfr,
    locations = ~abb
  ) %>% 
  layout(
    geo = list(
      scope = "usa",
      showlakes = TRUE,
      lakecolor = toRGB("red")
    )
  )